Begins at 9am JST / 4pm PT
Join Mark Chen, Josh Tobin, Neel Ajjarapu, and Isa Fulford from Tokyo as they introduce and demo deep research.
10 Likes
Coming to ChatGPT Pro tomorrow… Plus and others later…
Basically, multi-step web search that takes 5+ minutes…
5 Likes
https://openai.com/index/introducing-deep-research/
Summary of OpenAI’s Deep Research Announcement (February 2, 2025)
Introduction:
OpenAI introduced deep research, a new agent within ChatGPT that autonomously conducts multi-step research tasks on the internet. It processes extensive online information and synthesizes it into research reports at an analyst level.Key Features:
- Uses the upcoming OpenAI o3 model, optimized for web browsing and data analysis.
- Provides in-depth, well-documented reports with citations.
- Can analyze large amounts of text, images, and PDFs, pivoting as needed based on new findings.
- Designed for professionals in finance, science, policy, and engineering, as well as consumers seeking detailed product recommendations.
Functionality:
- Available to Pro users today; Plus and Team users will receive access next.
- Users can activate deep research within ChatGPT by selecting it in the message composer.
- Research tasks may take 5–30 minutes, and users receive notifications upon completion.
- Reports will soon include embedded images, data visualizations, and additional analytical outputs.
Performance & Benchmarks:
- Achieved 26.6% accuracy on Humanity’s Last Exam, surpassing previous AI models in expert-level knowledge tasks.
- Set a new state-of-the-art (SOTA) performance on GAIA, a benchmark for real-world AI questions.
- Demonstrated significant time savings, completing research tasks that would take humans multiple hours.
Limitations & Future Plans:
- May still hallucinate facts and struggle with uncertainty calibration.
- Currently compute-intensive, but a faster, cost-effective version is in development.
- Future updates will expand access to specialized and subscription-based data sources.
- OpenAI envisions integrating deep research with Operator for executing real-world tasks.
Deep research is now available on the ChatGPT web app, with rollout to mobile and desktop apps expected soon.
3 Likes
Showing an example with super-long and detailed output.
1 Like
Huh. Fascinating. This essentially seems to be the consumer’s first taste of adding test-time compute to their problems. Which I do like.
There’s a bunch of projects I’d love to try this on to assess its validity and usefulness.
I still don’t like the fact it’s locked under a $200 per month price tag until they decide to let it trickle down to us poors, but aw well I guess.
I’m getting worried OAI is going to soon add a $2000 per month subscription after they release o3 regular and then add the new and novel features/models to that subscription after they’re done with the $200 per month one.
I guess all we can do is see what people post about using these new agentic systems and hope for the best.
woo.
1 Like
I’m a Pro subscriber and don’t see the Deep Research option in my account yet. Is this slowly rolling out to people or has something else gone astray? Keen to try it out. I’m in Australia.
5 Likes
Also not seeing it yet in the States. Presumably a gradual rollout, but I hope to see it soon.
3 Likes
I also dont see it, keen to use it ASAP 
1 Like
As a GPT-pro user from Australia, I’m getting concerned that I keep hearing about all these new features that are “available to pro users” that I don’t see. Operator and now Deep Research are no-where to be seen and we’re paying the same as everyone else.
1 Like
100% there is so much searching Operator could be doing for me that takes up so much of my day
It landed sometime in the night/morning for me in the US. It’s absolutely amazing! Feels like that OMG-moment when first trying intelligent LLMs.
I don’t see it doesn’t mean that I don’t have access to it I’m on the G P T Pro